Solved – How to use Turnbull’s nonparametric estimator for interval-censored data in R

I am doing Survival Analysis with interval-censored data and I want to apply Turnbull's nonparametric estimator to the analysis of the covariates (suggested by Turnbull (1976)).

I want to know if there anyone knows any statistical package in R wich produces the survival curve estimate based on Turnbull´s algorithm.

If you want the Non-Parametric Maximum Likelihood Estimator (NPMLE) (what Turnbull 1976's algorithm computes), you could use the interval package.

I believe the default uses the standard EM algorithm as presented by Turnbull…but this is just about the slowest algorithm out there for computing the NPMLE. This could be a problem with n > 500, for example (also depends on the number of unique interval times). In interval, you can chose other faster algorithms with the initfit option; for example, I believe currently the fastest option would be to set initfit = 'initcomputeMLE'. The package also includes various plotting and testing functions.

Currently, the fastest algorithm available is ic_np in the icenReg package (note: this is the author). This can be fit as

npmle_fit <- ic_np(cbind(l, r) ~ 0, data = myData)

(where l and r are variables for the left and right side of the intervals in the dataset myData). And the plot can be produced by


However, this is currently not compatible with the interval functions, which can do a lot more with the NPMLE than just plot it. So unless speed is really a huge issue, you may want to stick with interval.

Similar Posts:

Rate this post

Leave a Comment